Fig. 1

SorpVision Dataset Creation for Automated Water Absorption Measurement in Cementitious Materials. (a) Orthogonally aligned USB microscope cameras capture liquid infiltration under LED-driven specular lighting (top) and ambient diffuse lighting (bottom) conditions. (b) Real and synthetic image generations, along with their binary masks. (c) Four annotators created binary masks of the water levels in paste samples. (d) Error maps reveal increased variability at lower water levels. (e) More annotators reduce variability, with diminishing discrepancies beyond three annotators. (f) At lower water levels, annotators produce higher error rates and variability, making it harder to capture fine details accurately. The data, adapted from Kabir et al.²⁰, is presented in a modified format.